Debayer algorithms - a comparison

[mcrossley]

I would like to have access to the Variable number of gradients algorithm since that looks the business.

I wrote a MaximDL demo plugin that implements the following CFA conversion algorithms:

- None (converts the CFA to RGB)

- Nearest Neighbour

- Bilinear

- Smooth Hue Transition

- Smooth Hue Transition Log

- Edge Sensing 1

- Edge Sensing 2

- Pattern Recognition

- Colour Correction 1

- Variable Number of Gradients

- Pixel Grouping

I have found VNG to be the most consistent, but different images work better with different algorithms. The end colour balance has nothing to do with the CFA conversion, it is background calibration and RGB weighting that affects that. Something that is fairly easily adjusted in PS.

Mark

[mcrossley]

The Maxim 'Quick' colour convert results in VERY poor star shapes, zoom in and check them.

Mark

[centroid]

Mark,

so is just by pure luck, that AA4 gets it "right" every time when debayering my H9C images, or is it as I'm thinking, thatMSB have set their debayering function 'tailor made' for SX camera images? :?

Dave

[centroid]

Here's another one to add to the 'equation'

This time, the same image debayered in Nebulosity 2, at its default settings.

Dave

[mcrossley]

...so is just by pure luck, that AA4 gets it "right" every time when debayering my H9C images...

No, along with CFA conversion, many programs can also perform automatic balancing - some more successfully than others - of both background levels and RGB weighting. Maxim allows you to do this automatically, or manually (or not at all). DSS for instance allows 'per channel' or 'RGB' calibration. Nearly all of them fail on images with extensive nebula over much of the background, they set the black point to what should be red nebula. All the automatic guestimates are just that - estimates, some closer than others, but nearly always you will have to make manual 'tweaks' to get it somewhere closer to 'true' colour. My Maxim plugin has the option to equalise the back levels of the channels, it just saves you a rough step in post processing. If anyone wants to play with the plug-in drop me a PM, though I may be offline for a couple of days whilst I digest the turkey!

Of the original images you posted, I suspect that JW's plugin produced closest to what the camera actually captured - quite a bit of red/orange light pollution in the background? By producing a black background the others have changed 'reality' to compensate for that red bias and produce a more pleasing image, which is easier to carry on post processing.

[MartinB]

Thanks for that Mark, very informative. PM sent!

[winfij]

Hi all,

Just found this thread.

I'd agree wholeheartedly with Mark - the key to analysing the performance of debayering algorithms isn't to look at the widescale color balance - that's easily (and arbitarily) changed, but rather to zoom right in and look at how they handle the nuances of changes from one color area to another - the edge of stars being especially tough due to the relatively sharp transition to black.

My plugins do indeed generally perform no automatic post-conversion conversion processing, so should give a direct representation of the color captured by the camera, which is almost never with a black sky (unless you live in the outback).

I did add options to perform an automatic background adjust and balance adjust for convenience, since the Maxim process offered them, but those are no different than the seperate tools already available in Maxim.

I was under the impression that the "high quality" conversion in Maxim now did VNG, from conversations I had with Doug George.

John

[mcrossley]

I was under the impression that the "high quality" conversion in Maxim now did VNG, from conversations I had with Doug George.

John, any idea when 'now' was from? Ver 4.6?

[winfij]

I think from the first release of the new algorithms, they went straight from nearest-neighbor to VNG - quite a change!

Doug told me he thought they were using bilinear and was a bit surprised when I showed him it was still nearest-neighbor

John

[centroid]

Hi John

Thanks for responding to this 'one'.

Unlike your goodself, and others, I have no in-depth knowledge of debaying algorithms, and only admiration for those able to adapt and write them into software packages.

As an 'end user', I can only report on what I find, and believe that my 'experiment' was carried out in a completely fair mannner, with the results being a true representation of what I obtained.

As I said, the same 'undoctored' sub, was debayered in each test, and no further processing applied.

Having been to a star party in the Aussie outback, I do appreciate what a truly dark sky looks like, and while my skies at home are not in the 'same league', I don't do too bad for light pollution.

I would accept that my skies are not truly black, but neither are they red.

Having conversed with 'the man' at CCDware, re CCDstack, I'm now aware that the debayering process used in CCDstack, applies no colour balancing, just a straight forward debayer function.

I believe, programs like AA4 and Maxim, do apply some colour balancing in the background, during debayer.

Dave